The network structure of our system has changed over the years. Broadly
speaking it has changed from a Mainframe system to a distributed,
In a mainframe system there's a central machine on which all the programs are run. In a typical distributed system, the programs are run on the machine the user is sat at - the users perhaps sharing resources like printers, etc. Each system has pros and cons
- A long time ago, the machine room was twice as big as it is today and
contained an IBM Mainframe and stored files and ran programs.
- That was replaced by an HP fileserver plus
HP terminals running HP-UX.
- The terminals proved too slow to run many programs,
so we got some "CPU servers" to run programs, using the terminals merely
to do input/output (i.e. they became "thin clients"). Each group of
terminals connected to a CPU server was called a "cluster" - a term
that we still use, though the meaning has changed.
- Small groups of Windows machines were added to the system - one at the end
of the DPO, for example.
- Meanwhile, a few infrastructure machines and some machines in the EIETL
were running Linux. In addition, Peter Long's MIT-related MDP project
resulted in a CD containing Linux and many CUED programs. For the 2005-6
year we offered an MDP-like option on the DPO terminals alongside access
to the HP-based system. For 2006-7 we removed the HP option.
- In 2008 we used SuSE Linux instead of the Knoppix-based MDP linux version.
- By 2011, all the DPO machines were "dual boot" (Windows or Linux)
- In 2012 we used Centos Linux instead of the SuSE linux version.
Many of the choices are cost-driven. Now that so many users carry their computers with them, and "cloud computing" is more common, the structure may be changing again. The moral of the story is - expect change!
Updated September 2010